class MyGraphCuts:
bgr_value = 0
obj_value = 1
none_value = 2
def __init__(self, img, sigma:float = 0.5, lam :float =1, gamma: float =0.5):
self.fig = Figure()
self.pres = GraphCutsPresenter(img, self)
self.pres.connect_figure(self.fig)
self.img = img
self.num_rows = img.shape[0]
self.num_cols = img.shape[1]
self.sigma = sigma
self.lam = lam
self.gamma = gamma
def run(self):
self.fig.show()
def compute_labels(self, seed_mask):
num_rows = self.num_rows
num_cols = self.num_cols
# +---------+---------+
# | | |
# | bgr | none |
# | | |
# +---------+---------+
# | | |
# | none | obj |
# | | |
# +---------+---------+
label_mask = np.full((num_rows, num_cols), self.none_value, dtype='uint8')
label_mask[:num_rows // 2, :num_cols // 2] = self.bgr_value
label_mask[num_rows // 2:, num_cols // 2:] = self.obj_value
return self.milestone_2(seed_mask) #label_mask
def milestone_1(self, seed_mask):
sigma = self.sigma
lam = self.lam
intensities = np.linalg.norm(x=self.img, axis=2)
right = np.square(intensities - np.roll(a=intensities, shift=-1, axis=0)) / (-2 * sigma ** 2)
below = np.square(intensities - np.roll(a=intensities, shift=-1, axis=1)) / (-2 * sigma ** 2)
# mask_right = np.where(seed_mask == np.roll(a=seed_mask, shift=-1, axis=0), 0, 1)
# mask_below = np.where(seed_mask == np.roll(a=seed_mask, shift=-1, axis=1), 0, 1)
# n_right = np.multiply(np.exp(right), mask_right)
# n_below = np.multiply(np.exp(below), mask_below)
n_right = np.exp(right) * lam
n_below = np.exp(below) * lam
t_source = np.where(seed_mask == self.bgr_value, np.infty, 0)
t_sink = np.where(seed_mask == self.obj_value, np.infty, 0)
g = maxflow.GraphFloat()
nodeids = g.add_grid_nodes((self.num_rows, self.num_cols))
structure_x = np.reshape(a=[0,0,0,0,0,1,0,0,0], newshape=(3,3))
structure_y = np.reshape(a=[0,0,0,0,0,0,0,1,0], newshape=(3,3))
g.add_grid_edges(nodeids, weights=n_right, structure=structure_x, symmetric=True)
g.add_grid_edges(nodeids, weights=n_below, structure=structure_y, symmetric=True)
g.add_grid_tedges(nodeids, t_source, t_sink)
g.maxflow()
seg = g.get_grid_segments(nodeids)
return np.array(seg, dtype=int)
def milestone_2(self, seed_mask):
sigma = self.sigma
lam = self.lam
gamma = self.gamma
intensities = np.linalg.norm(x=self.img, axis=2)
right = np.square(intensities - np.roll(a=intensities, shift=-1, axis=0)) / (-2 * sigma ** 2)
below = np.square(intensities - np.roll(a=intensities, shift=-1, axis=1)) / (-2 * sigma ** 2)
n_right = np.exp(right) * lam
n_below = np.exp(below) * lam
t_source = np.where(seed_mask == self.bgr_value, np.infty, 0) # bgr=0
t_sink = np.where(seed_mask == self.obj_value, np.infty, 0)
g = maxflow.GraphFloat()
nodeids = g.add_grid_nodes((self.num_rows, self.num_cols))
structure_x = np.reshape(a=[0,0,0,0,0,1,0,0,0], newshape=(3,3))
structure_y = np.reshape(a=[0,0,0,0,0,0,0,1,0], newshape=(3,3))
g.add_grid_edges(nodeids, weights=n_right, structure=structure_x, symmetric=True)
g.add_grid_edges(nodeids, weights=n_below, structure=structure_y, symmetric=True)
## T-Links
g.add_grid_tedges(nodeids, t_source, t_sink)
# Method 2
bgr_gmm = mixture.GaussianMixture(n_components=6)
obj_gmm = mixture.GaussianMixture(n_components=6)
bgr_mask = np.array(seed_mask == self.obj_value)
obj_mask = np.array(seed_mask == self.bgr_value)
if bgr_mask.sum() > 0 and obj_mask.sum() > 0:
bgr_samples = np.reshape(self.img[bgr_mask], newshape=(bgr_mask.sum(), self.img.shape[-1]))
obj_samples = np.reshape(self.img[obj_mask], newshape=(obj_mask.sum(), self.img.shape[-1]))
bgr_gmm.fit(bgr_samples)
obj_gmm.fit(obj_samples)
im_2d = np.reshape(self.img, newshape=(self.num_cols*self.num_rows, self.img.shape[-1]))
bgr_lhood = np.exp(bgr_gmm.score_samples(im_2d))
obj_lhood = np.exp(obj_gmm.score_samples(im_2d))
bgr_llhood = -np.log(gamma*bgr_lhood + (1-gamma)/256)
obj_llhood = -np.log(gamma*obj_lhood + (1-gamma)/256)
bgr_llhood_im = np.reshape(bgr_llhood, newshape=(self.num_rows, self.num_cols))
obj_llhood_im = np.reshape(obj_llhood, newshape=(self.num_rows, self.num_cols))
g.add_grid_tedges(nodeids, bgr_llhood_im, obj_llhood_im)
g.maxflow()
seg = g.get_grid_segments(nodeids)
return np.array(seg, dtype=int)
